Optimization of key parameters for airflow control and dust suppression using long-forced and short-exhaust ventilation at a fully mechanized heading face

To address the problems of complex coupling among key parameters and low dust reduction efficiency caused by experience-based regulation in the long-forced and short-exhaust airflow control and dust suppression system at fully mechanized heading faces in coal mines, a method for optimizing key parameters for airflow control and dust suppression using long-forced and short-exhaust ventilation at a fully mechanized heading face was proposed. In this method, a BP neural network and a Policy Gradient (PG) algorithm were used to construct a BP+PG fusion model. The BP neural network was used to establish the nonlinear mapping relationship between key dust suppression parameters, including exhaust air volume, radial-to-axial air volume ratio, and the distance between the dust control device and the working face, and normalized dust concentration. The PG algorithm enabled the model to adaptively adjust parameters in a continuous action space. On this basis, a fitness function with predicted dust concentration as the objective was constructed, and Particle Swarm Optimization (PSO) was used to globally optimize the key dust suppression parameters, thereby obtaining the optimal parameter combination. The comparison results of different models showed that, on the same test set, the BP+PG fusion model achieved a Mean Absolute Percentage Error (MAPE) of only 3.40% and a coefficient of determination R² of 0.97, representing significant improvements compared with the traditional BP neural network, BP+PPO, and BP+DQN. After the optimal parameter combination obtained using the proposed optimization method for key parameters of long-forced and short-exhaust airflow control and dust suppression was applied, namely an exhaust air volume of 450 m³/min, a radial-to-axial air volume ratio of 1.5, and a distance of 16 m between the dust control device and the working face, the measured dust concentration at the driver's position was 46.24 mg/m³, which was 63.2% lower than that before optimization (125.6 mg/m³), effectively improving the working environment in the driver's operating area.